Upload
hanga
View
270
Download
0
Embed Size (px)
Citation preview
Bitcoin Microstructure and the Kimchi premium
Kyoung Jin Choi
University of CalgaryHaskayne School of Business
Alfred Lehar
University of CalgaryHaskayne School of Business
Ryan Stauffer
University of CalgaryHaskayne School of Business
May 31, 2018
Haskayne School of Business, University of Calgary, 2500 University Drive NW, Calgary, Alberta, CanadaT2N 1N4. e-mail: [email protected]: [email protected]: [email protected]
Bitcoin Microstructure and the Kimchi premium
Abstract
Between January 2016 and February 2018, Bitcoin were in Korea on average 4.73%more expensive than in the United States, a fact commonly referred to as the Kimchi pre-mium. We argue that capital controls create frictions as well as amplify existing frictionsfrom the microstructure of the Bitcoin network that limit the ability of arbitrageurs to takeadvantage of persistent price differences. We find that the Bitcoin premia are positively re-lated to transaction costs, confirmation time in the blockchain, and to Bitcoin price volatilityin line with the idea that the delay and the associated price risk during the transaction periodmake trades less attractive for risk averse arbitrageurs and hence allow prices to diverge. Across country comparison shows that Bitcoin tend to trade at higher prices in countries withlower financial freedom. Finally unlike the prediction from the stock bubble literature, theKimchi premium is negatively related to the trading volume, which also suggests that theBitcoin microstructure is important to understand the Kimchi premium.
Keywords: Bitcoin, Limits to Arbitrage, Cryptocurrencies, Fintech
1 Introduction
I think the internet is going to be one of the major forces for reducing the role of
government. The one thing thats missing but that will soon be developed, is a
reliable e-cash.
Milton Friedman in 1999 - Nine years later, Bitcoin was created.
Many proponents of cryptocurrencies such as Bitcoin list independence from government
influence as key advantage of this new technology. In an ideal world, payments can be made and
funds exchanged globally without any central authority or government regulation.1 Yet we argue
in this paper that government regulations in fiat currencies, especially capital controls, create
new and amplify existing frictions in the global Bitcoin market. In Korea, for example, Bitcoin
frequently trade at a higher price than in other markets, a phenomenon referred to as the Kimchi
premium. Between Between January 2016 and February 2018 the average Kimchi premium was
4.73% but it reached levels as high as 54.48% in January 2018. Figure 1 shows a time series plot
as well as a histogram of the historical Kimchi premium. In frictionless financial markets such
a price difference could not persist as it would be immediately arbitraged away. Traders could
buy Bitcoin in another market, say the US, then transfer them to a Korean Bitcoin exchange,
sell them for Korean won, and convert the won to US dollars for an instant profit. However,
institutional frictions prevent arbitrageurs from keeping Bitcoin prices in Korea aligned with
the rest of the world. Divergence in Bitcoin prices are not only a Korean phenomenon. As we
document in this paper, international differences in Bitcoin prices can be high and persist over
longer periods of time. Even within the US prices differ somewhat between exchanges.2 In this
paper we analyze two main frictions that can contribute to a potential misalignment of Bitcoin
1Satoshi Nakamoto, a pseudonym for the legendary inventor of Bitcoin, included the headline of the FinancialTimes on Jan 3, 2009, Chancellor on brink of second bailout for banks in the first block, the genesis block, of theBitcoin Blockchain. Many see this as an expression of distrust in the current financial system. As the first reasonfor the existence of Bitcoin, Bitcoin-Wiki states that Bitcoin is P2P electronic cash that is valuable over legacysystems because of the monetary autonomy it brings to its users.
2Several websites allow users to monitor spreads and identify possible arbitrage opportunities: e.g.,www.tokenspread.com, data.bitcoinity.org/markets/arbitrage. Figure 4 in the appendix shows such an arbitragematrix.
1
prices across major markets: capital controls and frictions emanating from the microstructure
of the Bitcoin network.
Due to microstructure of the Bitcoin network arbitrageurs are confronted with obstacles that
are absent in many traditional markets. An arbitrageur faces two main sources of risk when exe-
cuting the arbitrage trade described above. First, the transfer of Bitcoin from a foreign exchange
to a Korean exchange takes time during which the Bitcoin price can change dramatically. Since
Bitcoin can usually not be shorted in the markets where they are selling at a higher price the
premium cannot be locked in; Bitcoin at a Korean exchange can only be sold once the transfer is
complete. Because Bitcoin can be much more volatile than many transitional assets, price risk
can pose a significant deterrent for arbitrageurs. Second, time varying transaction costs erode
potential arbitrage profits. Demand for transactions fluctuates over the day and over time. As
fees increase, profits from arbitrage decrease allowing the price difference between Bitcoin in
Korea and the rest of the world to rise.
Frictions in traditional capital markets add limitations to arbitrage. Korean capital controls
limit the amount of money that can be sent abroad, or at least complicate the transfer of funds,
and thus create a one way friction for the fiat currency part of the arbitrage trade. In the af-
termath of the global financial crisis and the European sovereign debt crisis, Korea introduced
capital controls that create administrative burden and additional time delay when sending money
abroad.
Our main finding is that both Bitcoin microstructure effects as well as exchange controls
explain a significant portion of Kimchi premium. We start out with an in-depth comparison
of the Korean Bitcoin market with the European market where more detailed data is available
and markets are well developed and liquid. We find that in both markets microstructure effects
are correlated with price deviations. Price deviations relative to the US market are significantly
positively related to Bitcoin volatility, supporting the idea that price risk for traders limits arbi-
trage activity. The Kimchi premium is also positively related to the median confirmation time
in the block chain, supporting the idea that longer transaction times create more uncertainty for
2
Figure 1. The Bitcoin Kimchi Premium: Bitcoin frequently trade at a higher price in Korea thanin other markets. The premium for purchasing Bitcoin with Korean won (KRW) versus US dollars(USD) is calculated: (KRWBTCprice in USD)/(USDBTCprice) 1, where the Bitcoin price in USD is themean price of all USD transactions on the Bitstamp exchange for that day. The Bitcoin price in KRW issimilarly defined from the Korbit exchange. Conversion from KRW to USD is done using the OANDAdaily average rate.
a) The Kimchi premium over time (top panel)
b) Distribution of the Kimchi premium (bottom panel)
3
arbitrageurs allowing prices to diverge. Finally price differences are also increase in transaction
fees paid to miners, consistent with the idea that higher fees reduce the attractiveness of the
arbitrage trade. In particular, if the trader offers higher fees, miners are more likely to include
the transaction in a block, hence making the transfer of Bitcoin faster. Therefore, the effect of
transaction fees should be inclusive of that of the median confirmation time, which is shown in
our regression analysis.
Comparing the European to the Korean market, however, we find two important differences.
First, the impact of microstructure effects on price divergences are several times larger for Korea
than for the European market. Second, we find the Kimchi premium to be positive while the
average premium for the European market is near zero. We argue that capital controls are the
reason why the average premium in positive and also make the the premium more sensitive to
microstructure effects. Because of the asymmetry of capital controls (it is easier to move funds
to Korea than the other way around) arbitrage is harder on one direction, allowing the Kimchi
premium to be positive on average. Because of the capital controls arbitrage is more costly and
hence the premium in Korea is more sensitive to transaction costs, volatility, and transaction
times compared to the European market.
To further analyze the impact of capital controls we collect data on Bitcoin premia for an
international sample and analyze how premia vary with various measures of financial freedom.
Controlling for microstructure effects we find that countries with higher financial freedom have
on average lower premia. Figure 2 plots the median Bitcoin premium from March 2017 to the
end of February 2018 as a function of financial freedom.3 As a stylized fact, the graph shows
higher average premia in financially more restrictive countries which is consistent with our view
that financial restrictions are causing higher Bitcoin prices in some countries.
We also examine Korean premia in other cryptocurrencies such as Ethereum, Lite-coin, and
Ripple. Instead of using fiat currency, arbitrageurs could complete the arbitrage trade by buying
other crypto currencies with the proceeds of selling Bitcoin in Korea and sending them abroad
3In This graph we use the Economic Freedom Ranking by the Fraser institute as explained in more detail inSection 5
4
Figure 2. Bitcoin Premia and Financial Freedom: The Bitcoin premium is measured as the medianpercentage price difference to the USD price from March 2017 to the end of February 2018. Bitcointransaction prices are from bitcoincharts.com, foreign exchange data from the Federal Reserve Bank ofSt. Louis (where available) and OANDA otherwise. To measure financial freedom we use the EconomicFreedom Ranking by the Fraser institute.
without being subject to capital controls. We find that other crypto currencies have practically
identical premia to Bitcoin at Korean exchanges and those premia are highly correlated over
time with the Kimchi premium.
In addition to our finding on the two main factors to derive the Kimchi premium, it is no-
table that the Kimchi premium is negatively related to the trading volume. This result sounds
quite counterintuitive in the sense that the trading volume is usually positively related to the
size of a bubble in common stock markets as shown by the traditional bubble literature (see,
e.g., Scheinkman and Xiong (2003) and Xiong and Yu (2011)). More precisely in our study,
the trading volume, with nothing else controlled for, is positively associated with the Kimchi
premium. However, if we add the trading volume into the regression together with volatility and
transaction fees, the Kimchi premium shows the negative relation with the trading volume while
5
its positive correlations with the other two factors become stronger. Our findings are consistent
with two countervailing forces. On one hand, a higher volume helps to reduce the Kimchi
premium through increased liquidity in the Korean Bitcoin market. On the other hand, high
volume increases blockchain transaction fees and blockchain confirmation times (and possibly
exchange cash-out times) and thus reduces the ability to arbitrage. Given that the volatility is ex-
tremely high, a high volume contributes more to increasing liquidity than increasing transaction
fees, which leads to the negative relationship between the premium and the volume. Therefore,
we argue that this negative relationship is attributed to the Bitcoin microstructure characteristics
that do not exist in a common stock market.
Our paper is related to the recent emerging and fast-growing literature on blockchain and
cryptocurrencies such as Athey, Parashkevov, Sarukkai, and Xia (2016), Cong, Li, and Wang
(2018), Detzel, Liu, Strauss, Zhou, and Zhu (2018), Pagnottayand and Buraschi (2018) and
Sockin and Xiong (2018), Griffin and Shams (2018). While most of these papers focus on
cryptocurrency pricing and/or asset pricing implications, we investigate the mechanism and the
factors that lead to the different pricing.4 Gandal, Hamrick, Moore, and Oberman (2018) inves-
tigate how suspicious trading activities caused a price spike in late 2013. Easley, OHara, and
Basu (2017), Huberman, Leshno, and Moallemi (2017), Cong, He, and Li (2018) investigate the
Bitcoin mining structure, competition of mining pools, and its impact on transaction fees. These
papers are important to understand speculative trading in the Bitcoin market and how Bitcoin
transaction fees are determined. Our paper adds to the literature by providing the explanation
of how these mechanisms generate pricing difference. In a contemporary paper Makarov and
Schoar (2018) analyze spreads in an international sample of Bitcoin prices are relate these to
signed order flow.
Our work is also related to a broad literature on bubbles and limits of arbitrage (see Xiong
(2013) and Gromb and Vayanos (2010) for a survey). There are various constraints and limi-
tations known in the literature to impede arbitrage trading. Among such constraints5, risk (the4Hu, Parlour, and Rajan (2018) provides several stylized facts on cryptocurreny markets as well as a survey of
the literature on Bitcoin valuation.5The constraints include information asymmetry, short-sale constraints, leverage margin constraints, constraints
6
price volatility in our case) and international trading frictions (the capital controls in our case)
seem the most relevant to explain Bitcoin price differences. Specifically, we find that the Kimchi
premium has a significant positive relation to Bitcoin price volatility and capital controls.6
There is one notable difference between our results and those from the traditional bubble
literature investigating the joint effect of short-sale restrictions and heterogeneous beliefs in the
stock market (e.g. Miller (1997), Harrison and Kreps (1978), Chen, Hong, and Stein (2002),
Scheinkman and Xiong (2003), Hong, Scheinkman, and Xiong (2006), and Mei, Scheinkman,
and Xiong (2009)). Under a short-sale constraint, the optimists are more likely to be marginal
buyers and the stock price tends to reflect optimists valuation more than that of pessimists,
which leads to a bubble (meaning a higher value than the fundamental value). This conventional
bubble literature predicts that the size of bubble is positively related to the volatility and the
trading volume. Our result on volatility is in line with the literature, but trading volume differs
from the previous literature in our case. There is short-sale restriction in the Bitcoin market. In
particular, Bitcoin futures were introduced in the U.S. in early 2018, but not in Korea at the time
of writing. In this sense, it might be natural to expect a positive relationship between the Kimchi
premium and Bitcoin trading volume in Korean exchanges. While the bubble argument would
suggest that volume and worldwide Bitcoin returns are positively correlated, it does not explain
the Kimchi premium. Therefore, we strongly believe that the negative correlation indicates the
importance of the Bitcoin microstructure for generating differential pricing, which is the crucial
difference between Bitcoin and common stock markets.
Our paper is also closely related to a long literature on internationally dual-listed stocks
(or Siamese twins) (see e.g., Rosenthal and Young (1990), Froot and Dabora (1999), De Jong,
Rosenthal, and Van Dijk (2009), Gagnon and Karolyi (2010) and references therein) document-
ing occurrences of significant price differences in those stocks listed in two countries. Each of
these papers finds that the price differences are attributed to tax, accounting, holding costs of
on equity capital, etc.6See, e.g. Edwards (1999) on the effectiveness of capital controls
7
the stock like idiosyncratic risk7, and so on. These factors, however, are unlikely to be relevant
to the Bitcoin market. Rather our findings indicate that the Bitcoin microstructure amplifies the
capital control effect on the price difference.
Finally, our paper is related to the purchasing power parity (PPP) puzzle literature in inter-
national economics.8 Viewing Bitcoin as an internationally traded asset (or commodity), we
can compare the Bitcoin price differential across countries with that of worldwide sold goods
such as McDonald Big Mac sandwich9 (Pakko and Pollard (2003) and Clements, Lan, and
Seah (2012)). Our dataset has the unique advantage that we can provide evidence at a much
higher frequency that much of the existing literature. There are several factors identified by
the PPP literature which lead to an international price difference: transportation costs, sales
(or value-added) taxes, trade restrictions (such as safety guard), domestic labor productivity
(when there are non-tradeable goods such as services), government expenditure, and current
account deficits. All of these factors are irrelevant to the Bitcoin market, except perhaps the
transportation cost which plays a similar role as the transaction fee in the Bitcoin market. In
the PPP case, the transportation costs always add nontrivial contributions to the PPP deviation.
However, it is notable that the transaction fee effect on the Bitcoin price difference becomes
negligible without capital controls as we find that the average Bitcoin price difference is near
zero between Europe and US, between which there are almost no capital controls. Therefore,
this discrepancy between the PPP case and the Bitcoin case also emphasizes the importance of
the capital controls contributing to the Kimchi premium.
The rest of the paper is organized as follows: Section 2 presents the institutional background
on Bitcoin microstructure and Korean capital controls, Section 3 details the data we use in
our analysis, in Section 4 we compare the Korean and the European market in detail, Section
5 expands our sample internationally, in Section 6 we examine other cryptocurrencies, and
Section 7 concludes.7The idiosyncratic risk means the risk unrelated to the risk of other securities in either of the competing markets8See, e.g., Rogoff (1996) and Taylor and Taylor (2004).9The Big Mac price index is annually published by Economist since 1986.
8
2 Institutional background
2.1 Bitcoin microstructure
The microstructure of Bitcoin markets stands out in many ways from traditional markets. Trans-
actions, i.e. the transfer from one wallet to another wallet, get posted within the Bitcoin peer-to-
peer network in the mem-pool, from where miners pick transactions to be mined into a block,
which gets then added to the blockchain. Many exchanges require a certain number of confir-
mations to credit Bitcoin that are transferred to the exchange to an account. A transaction with
n confirmations means that this transaction has been included in a mined block and that there
have been n 1 subsequent blocks mined in the blockchain. Time delay arises from the time
it takes for a transaction to be included in a mined block and from the time it takes to mine the
required number of subsequent blocks. The time to be included in a block, which is referred to
as confirmation time, can vary substantially. The historic maximum of the daily median confir-
mation time is 47 minutes, for our sample the daily median confirmation time is 11 minutes on
average.10 Note that the confirmation for individual transactions can differ substantially from
the median, especially when the tranmsactions offer low fees to the miners. The average time
between successfully mined blocks is 10 minutes.
Transaction fees are endogenously determined in the Bitcoin network. When posting a
transaction to the mem-pool the originator can set a fee that he or she is willing to pay to the
miner for the transaction to be included in the block. Miners can select transactions from the
pool and keep the fees upon successfully mining a block. Transactions with higher fees have
a higher probability to be included in a block. An arbitrageur thus faces a tradeoff between
offering a high fee that will get the transaction processed faster and mitigate price risk and the
cost of the higher fee which will directly reduce the arbitrageurs profit.
When trading on Bitcoin exchanges another layer of delay arises. Most exchanges offer
clients accounts similar to an account with a traditional stock broker. Trades are usually only
10source: https://blockchain.info/
9
possible between account holders at the same exchange and a trade is just recorded in the ledger
of the exchange, not on the blockchain. The Bitcoin transferred from the seller to the buyer are
held in the wallets of the exchange on the blockchain before and after the trade; the exchange
just records a change of ownership in its internal records. Account holders can request a trans-
fer to a private wallet out of the exchange account which will trigger a ledger entry on the
blockchain. While there is no data available on processing times by exchanges, anecdotal ev-
idence on several Bitcoin forums shows that processing times can be substantial with traders
waiting up to several days before exchanges transfer Bitcoin from their exchange-account to a
private wallet from which a transfer (to another exchange) can be initiated. In particular, 5-10
hours of processing time from a U.S. exchange to a Korean exchange is commonly reported
by major mass media in Korea. For example, Chosun Ilbo, a Korean newspaper, tested the
arbitrage and reported a processing time of 9 hours from Coinbase to Bithumb on December
26, 2017 when the Kimchi premium was about 28%.11 Also, the deposit and withdrawal of
fiat money can be subject to considerable delay. For example in Canada, processing times for
deposits and withdrawals can take up to several months. In part the delay is caused by banks
refusal to deal with cryptocurrency companies. Quadriga, one of the two established exchanges
in Canada has to rely on a Portuguese bank to process many of its fiat currency transfers.12 In
Figure 5 we present further anecdotal evidence from forum posts for Coinbase, a US exchange.
2.2 Capital Controls
On June 13, 2010, in the aftermath of the global financial crisis and the European sovereign
debt crisis, Korea introduced capital controls that were revised several times since. The Ko-
rean foreign exchange transaction law has been very restrictive. According to the most recent
11Chosun Ilbo (Daily Chosun) is the # 1 news paper company in South Korea in terms of the total number ofdaily printing. See the following news article by the Chosun Ilbo on January 4: http://news.chosun.com/site/data/html_dir/2018/01/04/2018010400441.html.
12See article I just want my money back. Couple had $100K wire stuck for months after trying to buy Bitcoin,GlobalNews, March 27, 2018.
10
law revision (valid since July 18th, 2017)13, an individual can send money up to 3,000 USD
per transfer and up to 20,000 USD in total between January 1st and December 31st through a
particular financial institution. The total maximum is limited to 50,000 USD a year through
different institutions.14 There are several alternative ways to send cash abroad. First, one can
use a Korean credit card when buying Bitcoin at an exchange in the US. However, the maxi-
mum amount of purchases outside of Korea is limited to 10,000 USD per year. In addition, this
transaction is considered as commodity purchase, which means the buyer should pay customs
on buying Bitcoin. One can send US dollars to someone (e.g., relatives or friends in the US)
who can help arbitrage trading through Paypal. In this case, however, Paypal automatically re-
ports this transaction to the US Internal Revenue Service (IRS) and the IRS normally considers
this money inflow to the receiver as taxable income if the transfer amount is sufficiently large
or the transfers occur on a regular basis. In addition, many Korean lawyers15 say that under the
current South Korean law it is not very clear if transferring Bitcoin between a Korean exchange
and exchanges in other countries is considered as capital in- and out-flow or commodity ex-
port/import. This legal interpretation issue might pose an additional risk since the government
might investigate transfer activities ex-post and accuse market participants of violation of the
law depending on how they interpret the law. In Figure 5 we present anecdotal evidence from
forum posts where people are actively seeking Korean partners for arbitrage trading. One post
suggests to meet in person at an airport to complete the arbitrage.
3 Data Sources and Model Variables
Bitcoin is very popular in Korea. As of February 1st, 2018, there are 16 cryptocurrency ex-
changes in South Korea. The five largest exchanges, in terms of trading volumes, are Upbit (#1
13See the government website on small foreign remitment: http://www.mosf.go.kr/nw/nes/detailNesDtaView.do?searchBbsId1=&searchNttId1=MOSF_000000000009556&menuNo=4010100.
14There are some exceptions. For example, the maximum per year is up to 100,000 USD for educational reasonssuch as tuition with proper evidence.
15See, e.g. http://hongbyun.tistory.com/22.
11
world ranking), Bithumb (#7), Coinone (#14), Korbit (#18), and Coinnest (#21).16 Korbit was
the first Korean Bitcoin exchange that opened in April, 2013. Then, Bithumb (January, 2014),
Coinone (August 2014), Coinnest (July 2017) and Upbit (October, 2017) followed. Until Upbit
started an exclusive partnership with Bittrex (a major U.S. based exchange) on October 2017,
Bithumb, Coinone, and Korbit had been the three major exchanges.17
Our primary variable of interest is the Bitcoin premium in local currency over the Bitcoin
price in USD. For Korea the KRW Bitcoin premium over USD, the Kimchi premium, is defined
as
PremiumKRW =KRW/BTC price USD/KRW exchange rate USD/BTC price
USD/BTC price(1)
The premium for the European market is defined similarly based on EUR prices.
For daily Bitcoin prices in USD, KRW, and EUR we look at all transactions on specific
Bitcoin exchanges (data accessed via bitcoincharts.com). Exchanges were selected due to data
availability, length of trading history, and both current and historical market share. USD data is
from Bitstamp. Bitstamp has offices in Luxembourg, London, and Berkeley. They are currently
the 3rd largest exchange for USD trades by volume and have the longest trading history of the
current major players. In the early days of Bitcoin trading the USD leader was Mt. Gox which
famously went bankrupt following a security breach. The sample used for regressions runs
from 2016-01-01 through 2018-02-28 and contains 13,151,367 total trades. The total notional
value (valued at the time of each trade) is USD 33.9b. KRW data is from Korbit. Korbit was
South Koreas first Bitcoin exchange. Korbit is fourth by volume for KRWBTC as of February
1, 2018, but has been second or third during the majority of the sample period. The sample
contains 4,889,844 total trades with a total notional value (at the time of each trade) of KRW
14.2t. EUR data is from Kraken. Kraken is currently the largest exchange for EURBTC by
volume, with more than half the total volume. The sample contains 15,349,375 trades with a
16The number inside the parenthesis is the world ranking in trading volumes (all the cryptocurrencies) by Coin-hills on February 1st, 2018 (see https://www.coinhills.com/market/exchange/.
17Among the top three, Korbit is the only one which provides a history of all the trades in unix-time
12
total notional value (valued at the time of each trade) of EUR 19.1b.
The daily USD price we utilize for analysis is the mean price of all USD transactions on the
Bitstamp exchange for that day. The KRW and EUR daily prices are similarly defined using
Korbit and Kraken exchanges, respectively. To convert the KRW and EUR prices to USD we
utilize data from OANDA. The daily prices utilized are the average price (not the close) over
the 24-hour period (UTC time standard) aggregated from multiple exchanges. We find this the
best fit for our purpose as the Bitcoin markets operate 24/7.
We estimate short term volatility for Bitcoin prices as the sum of 10 minute squared log-
returns over one day. Microstructure noise can arise from spreads between bid and ask prices
and from shifts in transaction prices due to the random execution of large trades at either end of
the 10 minute interval. We take two measures to mitigate potential biases due to microstructure
noise. First, we compute daily volatility for a given exchange as the average of two volatility
measures, based on 10 minute returns shifting the time interval by 5 minutes. Second, we
define volatility as the median volatility over several exchanges.18 To test for robustness we
also compute long term volatility for a given exchange as the sum of squared 12 hour returns
over a period of 20 days. We then define long term volatility as the median volatility over
several exchanges. Our results are robust with respect to this alternative volatility measure.
The Bitcoin blockchain median confirmation time data is from www.blockchain.info. This
is the median time in minutes for a Bitcoin transaction to be accepted into a mined block and
added to the public ledger; this number is computed only from transactions with miner fees and
might therefore underestimate actual confirmation times. For days with missing data (of which
there are none in the most recent 2 years) we interpolate linear between days. The maximum
gap in the data set was 1 day. Results were unchanged when using the previous days value or
removing missing days completely from analysis.
The mean blockchain transaction fee is measured in USD and calculated from data from18Data availability differs per time period as data is not available for all exchanges at all times. We include data
from the following exchanges: bitfinex, Bitstamp, BTCC, btc-e, coinbase, Gemini, hitbtc, itbit, kraken, OK-Coin,Poloniex as available on bitcoinchain.com.
13
blockchain.info. It is the total value of all transaction fees paid to miners converted to USD
(not including the value of block rewards), divided by the number of daily confirmed Bitcoin
transactions on the blockchain for that day.
For the KRW and EUR volumes we look at the daily total number of exchange transactions
(in thousands) on Korbit and Kraken respectively. This approach was taken rather than volume
in Bitcoin due to the wildly differing Bitcoin prices at the start versus the end of the sample
period. All results are robust to using an alternative volume measure of the daily USD equivalent
total valuation of transactions. For the KRW-USD and EUR-USD foreign exchange volatilities
we use the standard deviation of 1-day logarithmic returns in the daily average KRW-USD and
EUR-USD exchanges from OANDA, over the most recent 20 days.
The 1-day Bitcoin return variable is the 1-day logarithmic return in the USD-BTC price,
where USD-BTC daily price is calculated as described above for the Kimchi premium calcula-
tion. The FOMC week variable is a dummy variable which takes a value of 1 if that day is within
3 days of a US Federal Reserve Federal Open Market Committee (FOMC) meeting start date
(one week centered on the start date of the FOMC meeting). Results are unchanged if a leading
week is used instead of centered, and all FOMC meetings during the regression sample period
were regularly scheduled (the last unscheduled meeting was in March 2014). The FOMC sets
monetary policy for the US (including a target for the overnight interbank interest rate). Cor-
bet, Larkin, Lucey, Meegan, and Yarovaya (2017) found that certain classes of digital assets
(currencies, mineable), of which Bitcoin is a member, have volatility spillover transfer from
US monetary policy announcements. There is also a broad literature on the impact of FOMC
meetings/announcements on the volatilty, returns, and trading volume of other assets both in the
US and globally. For example, Fischer and Ranaldo (2011) found that foreign exchange trading
volume was significantly increased on FOMC days, and Lucca and Moench (2015) document
large average excess returns on US equities ahead of the monetary policy decisions made at
scheduled FOMC meetings. The Bitcoin news in Korea variable is the total number of daily
news articles published in Korea with keyword Bitcoin from Factiva.19 A summary of the
19Bitcoin is . in Korean.
14
variables used in our empirical analysis can be found in Table 8 in the appendix.
4 Empirical Results
To analyze the determinants of the Kimchi premium we regress daily observations of the relative
price difference for Bitcoin in Korea and the US on several factors or proxies for potential
frictions inhibiting the arbitrage. All Bitcoin for fiat currency transaction times are converted
to UTC time standard. Days with missing trading data are excluded. All results were robust to
testing on a sample with linear interpolation between missing days.
Regression results are shown in Table 1. Bitcoin price risk is a significant component to
the Kimchi premium size. In periods of high volatility, the cost of waiting for blockchain
confirmations could be very significant and deter arbitrageurs. Model (1) documents a positive
relation between the Kimchi premium and short term BTC volatility. Model (2) shows that
higher fees make the arbitrage less profitable and thus coincide with higher premia. As shown in
model (3) higher median confirmation times on the block chain are also associated with higher
Bitcoin premia. An arbitrageur could potentially jump the queue get her transaction processed
faster by offering a higher transaction fee to miners, yet such a higher transaction fee would also
cause a direct reduction in arbitrage profits and hence allow for a larger premium. Model (4)
shows that higher volume by itself is associated with higher fees. Considering all these factors
we document in model (5) that only short term volatility and transaction fees stay significantly
positive, while volume turns negative. Median confirmation time seems to be a similar proxy for
bottlenecks in the blockchain to transaction fees and becomes insignificant. FX-volatility is not
a driving factor behind the Kimchi premium as FX-volatility is substantially smaller than BTC
volatility. Higher volume may help reduce the Kimchi premium through increased liquidity in
the Korean Bitcoin market, while at the same time increase blockchain transaction fees and/or
blockchain confirmation times (and potentially exchange cash-out times) thus reducing ability
to arbitrage and increasing the Kimchi premium. In model (6) we find that a one day lagged
15
Tabl
e1:
Reg
ress
ion
resu
ltsfo
rth
eK
RW
Bitc
oin
prem
ium
over
USD
.Dai
lytim
ese
ries
regr
essi
ons:
the
depe
nden
tvar
iabl
eis
the
prem
ium
forp
urch
asin
gB
itcoi
nw
ithK
orea
nw
on(K
RW
)ver
susU
Sdo
llars
(USD
)and
isca
lcul
ated
:(K
RW
BT
Cpr
ice
inU
SD)/(U
SDB
TC
pric
e)1,
whe
reth
eB
itcoi
npr
ice
inU
SDis
the
mea
npr
ice
ofal
lUSD
tran
sact
ions
onth
eB
itsta
mp
exch
ange
fort
hatd
ay.T
heB
itcoi
npr
ice
inK
RW
issi
mila
rly
defin
edw
ithda
tafr
omth
eK
orbi
texc
hang
e.C
onve
rsio
nfr
omK
RW
toU
SDis
done
usin
gth
eO
AN
DA
daily
aver
age
rate
.T
hein
depe
nden
tva
riab
les
are
defin
edas
inTa
ble
8. (1)
(2)
(3)
(4)
(5)
(6)
(7)
Bitc
oin
shor
tter
m0.0667
0.0348
0.0374
0.0436
vola
tility
(0.0041)
(0.0047)
(0.0049)
(0.0051)
Mea
nbl
ockc
hain
0.0069
0.0065
0.0066
0.0069
tran
sact
ion
fee
(0.0003)
(0.0004)
(0.0004)
(0.0004)
Blo
ckch
ain
med
ian
0.0027
0.0006
0.0005
0.0004
confi
rmat
ion
time
(0.0007)
(0.0006)
(0.0006)
(0.0006)
KR
W-B
TC
volu
me
0.0045
0.0014
0.0017
0.0018
(tho
usan
dsof
tran
sact
ions
)(0.0003)
(0.0004)
(0.0004)
(0.0004)
KR
W-U
SD0.4514
0.5734
0.8579
vola
tility
(1.9711)
(1.9664)
(1.9429)
Bitc
oin
1-da
y0.1353
0.1249
lagg
edre
turn
(0.0589)
(0.0581)
FOM
Cw
eek
0.0182
(0.0055)
BT
Cne
ws
Kor
ea0.0001
(0.00003)
Con
stan
t0.0008
0.0242
0.0172
0.0195
0.0046
0.0049
0.0078
(0.0038)
(0.0022)
(0.0087)
(0.0030)
(0.0115)
(0.0115)
(0.0114)
Obs
erva
tions
788
790
790
790
788
788
788
R2
0.25
000.
4525
0.01
660.
2261
0.49
010.
4935
0.50
78
Not
e: p